Reliability Scales Inversely: Bigger Language Models Compound Mistakes Faster

arXiv:2607.18292v2 Announce Type: replace-cross Abstract: As language models scale, answers start truer but degrade faster: scaling buys capability but erodes reliability. The knowledge-gap account -- more data, retrieval, or scale -- misses an auto-regressive risk residual that increases with scale: the model commits to a low-probability token, conditions on it as established, and snowballs. We track this through per-position disagreement elta = log p_M - log p_O against a stronger same-family oracle, whose second moment splits exactly into bias^2 KL(p_M ,|, p_O)^2 and risk Var[elta]. Across three model families, we present four findings: (i) under scaling, the knowledge gap falls up to 7imes while knowledge degradation grows up to 39imes; (ii) at a fabrication, felt uncertainty H(p_M) relaxes quickly while oracle-referenced risk persists up to 23imes longer, leaving a confident-but-precarious risk regime that bridges consecutive fabrications; (iii) this regime is causal -- a fixed-KL variance contraction cuts web-verified hallucination 35-74%; and, (iv) it evades self-monitoring, with p_M-only detectors (e.g. semantic entropy) firing approx30% less (p!<!10^{-16}) on the risky branch despite it holding nearly 4imes more fabrications. Bigger models snowball mistakes faster, through a failure mode that is dominant, self-perpetuating, causal and invisible to the model itself.

Related

• RL-PLUS: Countering Capability Boundary Collapse of LLMs in Reinforcement Learning with Hybrid-policy Optimization
• Learning-Based Automated Adversarial Red-Teaming for Robustness Evaluation of Large Language Models
• Logit-Gap Steering: A Forward-Pass Diagnostic for Alignment Robustness

Source: arXiv cs.CL | 2026-07-27